Trending topics
#
Bonk Eco continues to show strength amid $USELESS rally
#
Pump.fun to raise $1B token sale, traders speculating on airdrop
#
Boop.Fun leading the way with a new launchpad on Solana.
AI Agent Context Never Lost: LCM Plugin's DAG Memory Architecture
Native OpenClaw (and almost all AI Agents) simply truncate old messages when conversations exceed the model's context window - information is lost as a result.
The core claim of Lossless Claw is: compression does not equal forgetting. It replaces the original sliding window truncation mechanism with a layered summary system based on a DAG (Directed Acyclic Graph), persistently storing each message, and through recursive condensation of summaries, allows the Agent to theoretically "remember" an infinitely long history while maintaining token budget.
• GitHub has gained 2k Stars, 147 Forks, and quickly became popular shortly after launch - a representative project among OpenClaw ecosystem tools.
• The threshold for context-triggered compression is set at 75% (contextThreshold=0.75), meaning condensation begins when there is still 25% headroom, avoiding overflow.
• Protects the latest 32 messages from being compressed (freshTailCount=32), ensuring recent coherence.
• Uses SQLite to persist all original messages, with summary nodes linked back to the source messages, allowing for recovery of the original text at any time.
• Provides three Agent tools: lcm_grep (search), lcm_describe (describe nodes), lcm_expand (expand details).
• Each leaf node can contain up to 20,000 tokens of source content, with a compression target of 1,200 tokens; higher-order condensed nodes target 2,000 tokens.
1. Installation: One command openclaw plugins install @martian-engineering/lossless-claw, no manual JSON modification required.
2. Configuration: Specify contextEngine: "lossless-claw" in the OpenClaw config, fine-tune parameters through environment variables.
3. Core Design Patterns:
• Automatic compression after each conversation round (can be turned off).
• Old messages → Leaf Summary → Condensed Node, condensing layer by layer to form a DAG.
• When the Agent calls lcm_expand, it unfolds from the DAG in reverse, restoring original details.
4. Session Persistence: Coupled with session.reset.idleMinutes: 10080 (7 days), allows the same session to last a week, with LCM memory accumulating across sessions....
Top
Ranking
Favorites
